Contents

Deep Learning: A Simple Example

  • Let’s get back to the Name Gender Classifier.

Prepare Data

import numpy as np
import nltk
from nltk.corpus import names
import random

labeled_names = ([(name, 1) for name in names.words('male.txt')] +
                 [(name, 0) for name in names.words('female.txt')])
random.shuffle(labeled_names)

Train-Test Split

from sklearn.model_selection import train_test_split
train_set, test_set = train_test_split(labeled_names,
                                       test_size=0.2,
                                       random_state=42)
print(len(train_set), len(test_set))

6355 1589

import tensorflow as tf
import tensorflow.keras as keras
from keras.preprocessing.text import Tokenizer
from keras.preprocessing import sequence
from keras.utils import to_categorical, plot_model
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import LSTM, RNN, GRU
from keras.layers import Embedding
from keras.layers import SpatialDropout1D

names = [n for (n, l) in train_set]
labels = [l for (n, l) in train_set]

len(names)

6355

Tokenizer

  • By default, the token index 0 is reserved for padding token.

  • If oov_token is specified, it is default to index 1.

  • Specify num_words for tokenizer to include only top N words in the model

  • Tokenizer will automatically remove puntuations.

  • Tokenizer use whitespace as word delimiter.

  • If every character is treated as a token, specify char_level=True.

tokenizer = Tokenizer(char_level=True)
tokenizer.fit_on_texts(names)

Prepare Input and Output Tensors

  • Like in feature-based machine translation, a computational model only accepts numeric values. It is necessary to convert raw text to numeric tensor for neural network.

  • After we create the Tokenizer, we use the Tokenizer to perform text vectorization, i.e., converting texts into tensors.

  • In deep learning, words or characters are automatically converted into numeric representations.

  • In other words, the feature engineering step is fully automatic.

Two Ways of Text Vectorization

  • Texts to Sequences: Integer encoding of tokens in texts and learn token embeddings

  • Texts to Matrix: One-hot encoding of texts (similar to bag-of-words model)

Method 1: Text to Sequences

From Texts and Sequences

  • Text to Sequences

  • Padding to uniform lengths for each text

names_ints = tokenizer.texts_to_sequences(names)

print(names[:10])
print(names_ints[:10])
print(labels[:10])

['Ivy', 'Forrest', 'Trula', 'Vail', 'Joelynn', 'Yard', 'Krysta', 'Melamie', 'Mikhail', 'Leone']
[[3, 20, 11], [21, 7, 5, 5, 2, 9, 8], [8, 5, 16, 6, 1], [20, 1, 3, 6], [19, 7, 2, 6, 11, 4, 4], [11, 1, 5, 10], [18, 5, 11, 9, 8, 1], [12, 2, 6, 1, 12, 3, 2], [12, 3, 18, 13, 1, 3, 6], [6, 2, 7, 4, 2]]
[0, 1, 0, 1, 0, 1, 0, 0, 1, 0]

Vocabulary

# determine the vocabulary size
vocab_size = len(tokenizer.word_index) + 1
print('Vocabulary Size: %d' % vocab_size)

Vocabulary Size: 30

tokenizer.word_index

{'a': 1,
 'e': 2,
 'i': 3,
 'n': 4,
 'r': 5,
 'l': 6,
 'o': 7,
 't': 8,
 's': 9,
 'd': 10,
 'y': 11,
 'm': 12,
 'h': 13,
 'c': 14,
 'b': 15,
 'u': 16,
 'g': 17,
 'k': 18,
 'j': 19,
 'v': 20,
 'f': 21,
 'p': 22,
 'w': 23,
 'z': 24,
 'q': 25,
 'x': 26,
 '-': 27,
 ' ': 28,
 "'": 29}

Padding

  • When padding the all texts into uniform lengths, consider whether to Pre-padding or removing values from the beginning of the sequence (i.e., pre) or the other way (post).

  • Check padding and truncating parameters in pad_sequences

names_lens = [len(n) for n in names_ints]
names_lens
import seaborn as sns
sns.displot(names_lens)
print(names[np.argmax(names_lens)])  # longest name

Helen-Elizabeth
../_images/dl-simple-case_25_1.png 
max_len = names_lens[np.argmax(names_lens)]
max_len

15

names_ints_pad = sequence.pad_sequences(names_ints, maxlen=max_len)
names_ints_pad[:10]

array([[ 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  3, 20, 11],
       [ 0,  0,  0,  0,  0,  0,  0,  0, 21,  7,  5,  5,  2,  9,  8],
       [ 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  8,  5, 16,  6,  1],
       [ 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0, 20,  1,  3,  6],
       [ 0,  0,  0,  0,  0,  0,  0,  0, 19,  7,  2,  6, 11,  4,  4],
       [ 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0, 11,  1,  5, 10],
       [ 0,  0,  0,  0,  0,  0,  0,  0,  0, 18,  5, 11,  9,  8,  1],
       [ 0,  0,  0,  0,  0,  0,  0,  0, 12,  2,  6,  1, 12,  3,  2],
       [ 0,  0,  0,  0,  0,  0,  0,  0, 12,  3, 18, 13,  1,  3,  6],
       [ 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  6,  2,  7,  4,  2]],
      dtype=int32)

Define X and Y

X_train = np.array(names_ints_pad).astype('int32')
y_train = np.array(labels)

X_test = np.array(
    sequence.pad_sequences(tokenizer.texts_to_sequences(
        [n for (n, l) in test_set]),
                           maxlen=max_len)).astype('int32')
y_test = np.array([l for (n, l) in test_set])

X_test_texts = [n for (n, l) in test_set]

print(X_train.shape)
print(y_train.shape)
print(X_test.shape)
print(y_test.shape)

(6355, 15)
(6355,)
(1589, 15)
(1589,)

Method 2: Text to Matrix

One-Hot Encoding

  • Text to Matrix (to create bag-of-word representation of each text)

  • Choose modes: binary, count, or tfidf

names_matrix = tokenizer.texts_to_matrix(names, mode="binary")

names[2]

'Trula'

  • names_matrix in fact is a bag-of-characters representation of a name text.

import pandas as pd
pd.DataFrame(names_matrix[2, 1:],
             columns=["ONE-HOT"],
             index=list(tokenizer.word_index.keys()))
ONE-HOT
a 1.0
e 0.0
i 0.0
n 0.0
r 1.0
l 1.0
o 0.0
t 1.0
s 0.0
d 0.0
y 0.0
m 0.0
h 0.0
c 0.0
b 0.0
u 1.0
g 0.0
k 0.0
j 0.0
v 0.0
f 0.0
p 0.0
w 0.0
z 0.0
q 0.0
x 0.0
- 0.0
0.0
' 0.0

Define X and Y

X_train2 = np.array(names_matrix).astype('int32')
y_train2 = np.array(labels)

X_test2 = tokenizer.texts_to_matrix([n for (n, l) in test_set],
                                    mode="binary").astype('int32')
y_test2 = np.array([l for (n, l) in test_set])

X_test2_texts = [n for (n, l) in test_set]

print(X_train2.shape)
print(y_train2.shape)
print(X_test2.shape)
print(y_test2.shape)

(6355, 30)
(6355,)
(1589, 30)
(1589,)

Model Definition

  • Three important steps for building a deep neural network:

    • Define the model structure

    • Compile the model

    • Fit the model

  • After we have defined our input and output tensors (X and y), we can define the architecture of our neural network model.

  • For the two ways of name vectorized representations, we try two different network structures.

    • Text to Matrix: Fully connected Dense Layers

    • Text to Sequences: Embedding + RNN

import matplotlib.pyplot as plt
import matplotlib
import pandas as pd


# Plotting results
def plot1(history):

    matplotlib.rcParams['figure.dpi'] = 100
    acc = history.history['accuracy']
    val_acc = history.history['val_accuracy']
    loss = history.history['loss']
    val_loss = history.history['val_loss']

    epochs = range(1, len(acc) + 1)
    ## Accuracy plot
    plt.plot(epochs, acc, 'bo', label='Training acc')
    plt.plot(epochs, val_acc, 'b', label='Validation acc')
    plt.title('Training and validation accuracy')
    plt.legend()
    ## Loss plot
    plt.figure()

    plt.plot(epochs, loss, 'bo', label='Training loss')
    plt.plot(epochs, val_loss, 'b', label='Validation loss')
    plt.title('Training and validation loss')
    plt.legend()
    plt.show()


def plot2(history):
    pd.DataFrame(history.history).plot(figsize=(8, 5))
    plt.grid(True)
    #plt.gca().set_ylim(0,1)
    plt.show()

Model 1: Fully Connected Dense Layers

  • Let’s try a simple neural network with two fully-connected dense layers with the Text-to-Matrix inputs.

  • That is, the input of this model is the bag-of-words representation of the entire name.

Dense Layer Operation

  • The transformation of each Dense layer will transform the input tensor into a tensor whose dimension size is the same as the node number of the Dense layer.

## Define Model
from keras import layers
model1 = keras.Sequential()
model1.add(keras.Input(shape=(vocab_size, ), name="one_hot_input"))
model1.add(layers.Dense(16, activation="relu", name="dense_layer_1"))
model1.add(layers.Dense(16, activation="relu", name="dense_layer_2"))
model1.add(layers.Dense(1, activation="sigmoid", name="output"))

## Compile Model
model1.compile(loss=keras.losses.BinaryCrossentropy(),
               optimizer=keras.optimizers.Adam(lr=0.001),
               metrics=["accuracy"])

plot_model(model1, show_shapes=True)
../_images/dl-simple-case_50_0.png

A few hyperparameters for network training

  • Batch Size: The number of inputs needed per update of the model parameter (gradient descent)

  • Epoch: How many iterations needed for training

  • Validation Split Ratio: Proportion of validation and training data split

## Hyperparameters
BATCH_SIZE = 128
EPOCHS = 20
VALIDATION_SPLIT = 0.2

## Fit the model
history1 = model1.fit(X_train2,
                      y_train2,
                      batch_size=BATCH_SIZE,
                      epochs=EPOCHS,
                      verbose=2,
                      validation_split=VALIDATION_SPLIT)

Epoch 1/20
40/40 - 2s - loss: 0.6730 - accuracy: 0.5901 - val_loss: 0.6497 - val_accuracy: 0.6404
Epoch 2/20
40/40 - 0s - loss: 0.6501 - accuracy: 0.6316 - val_loss: 0.6368 - val_accuracy: 0.6444
Epoch 3/20
40/40 - 0s - loss: 0.6360 - accuracy: 0.6459 - val_loss: 0.6235 - val_accuracy: 0.6617
Epoch 4/20
40/40 - 0s - loss: 0.6212 - accuracy: 0.6631 - val_loss: 0.6105 - val_accuracy: 0.6719
Epoch 5/20
40/40 - 0s - loss: 0.6077 - accuracy: 0.6703 - val_loss: 0.6019 - val_accuracy: 0.6782
Epoch 6/20
40/40 - 0s - loss: 0.5971 - accuracy: 0.6835 - val_loss: 0.5956 - val_accuracy: 0.6790
Epoch 7/20
40/40 - 0s - loss: 0.5892 - accuracy: 0.6888 - val_loss: 0.5895 - val_accuracy: 0.6861
Epoch 8/20
40/40 - 0s - loss: 0.5828 - accuracy: 0.6916 - val_loss: 0.5871 - val_accuracy: 0.6869
Epoch 9/20
40/40 - 0s - loss: 0.5775 - accuracy: 0.7022 - val_loss: 0.5822 - val_accuracy: 0.6932
Epoch 10/20
40/40 - 0s - loss: 0.5726 - accuracy: 0.7057 - val_loss: 0.5793 - val_accuracy: 0.6939
Epoch 11/20
40/40 - 0s - loss: 0.5690 - accuracy: 0.7063 - val_loss: 0.5779 - val_accuracy: 0.6987
Epoch 12/20
40/40 - 0s - loss: 0.5656 - accuracy: 0.7071 - val_loss: 0.5762 - val_accuracy: 0.7034
Epoch 13/20
40/40 - 0s - loss: 0.5630 - accuracy: 0.7111 - val_loss: 0.5782 - val_accuracy: 0.6979
Epoch 14/20
40/40 - 0s - loss: 0.5612 - accuracy: 0.7111 - val_loss: 0.5745 - val_accuracy: 0.7042
Epoch 15/20
40/40 - 0s - loss: 0.5580 - accuracy: 0.7144 - val_loss: 0.5747 - val_accuracy: 0.6987
Epoch 16/20
40/40 - 0s - loss: 0.5565 - accuracy: 0.7166 - val_loss: 0.5720 - val_accuracy: 0.7034
Epoch 17/20
40/40 - 0s - loss: 0.5545 - accuracy: 0.7179 - val_loss: 0.5717 - val_accuracy: 0.7050
Epoch 18/20
40/40 - 0s - loss: 0.5530 - accuracy: 0.7181 - val_loss: 0.5709 - val_accuracy: 0.7065
Epoch 19/20
40/40 - 0s - loss: 0.5518 - accuracy: 0.7209 - val_loss: 0.5707 - val_accuracy: 0.7081
Epoch 20/20
40/40 - 0s - loss: 0.5498 - accuracy: 0.7260 - val_loss: 0.5686 - val_accuracy: 0.7026

plot1(history1)
../_images/dl-simple-case_54_0.png ../_images/dl-simple-case_54_1.png 
model1.evaluate(X_test2, y_test2, batch_size=128, verbose=2)

13/13 - 0s - loss: 0.5687 - accuracy: 0.7042

[0.5686534643173218, 0.704216480255127]

Model 2: Embedding + RNN

  • Another possibility is to introduce an embedding layer in the network, which transforms each character in the name into a tensor (i.e., embeddings), and then to add a Recurrent Neural Network layer to process each character sequentially.

  • The strength of the RNN is that it iterates over the timesteps of a sequence, while maintaining an internal state that encodes information about the timesteps it has seen so far.

  • It is posited that after the RNN iterates through the entire sequence, it keeps important information of all previously iterated tokens for further operation.

  • The input of this network is a padded sequence of the original text (name).

Embedding Layer Operation

RNN Layer Operation

RNN Layer Operation

Unrolled Version of RNN Operation

Unrolled Version of RNN Operation

## Define the embedding dimension
EMBEDDING_DIM = 128

## Define model
model2 = Sequential()
model2.add(
    Embedding(input_dim=vocab_size,
              output_dim=EMBEDDING_DIM,
              input_length=max_len,
              mask_zero=True))
model2.add(layers.SimpleRNN(16, activation="relu", name="RNN_layer"))
model2.add(Dense(16, activation="relu", name="dense_layer"))
model2.add(Dense(1, activation="sigmoid", name="output"))

model2.compile(loss=keras.losses.BinaryCrossentropy(),
               optimizer=keras.optimizers.Adam(lr=0.001),
               metrics=["accuracy"])

plot_model(model2, show_shapes=True)
../_images/dl-simple-case_66_0.png 
history2 = model2.fit(X_train,
                      y_train,
                      batch_size=BATCH_SIZE,
                      epochs=EPOCHS,
                      verbose=2,
                      validation_split=VALIDATION_SPLIT)

Epoch 1/20
40/40 - 2s - loss: 0.6301 - accuracy: 0.6849 - val_loss: 0.5575 - val_accuracy: 0.7278
Epoch 2/20
40/40 - 0s - loss: 0.4917 - accuracy: 0.7537 - val_loss: 0.4768 - val_accuracy: 0.7569
Epoch 3/20
40/40 - 0s - loss: 0.4363 - accuracy: 0.7825 - val_loss: 0.4526 - val_accuracy: 0.7632
Epoch 4/20
40/40 - 0s - loss: 0.4224 - accuracy: 0.7925 - val_loss: 0.4414 - val_accuracy: 0.7718
Epoch 5/20
40/40 - 0s - loss: 0.4117 - accuracy: 0.8021 - val_loss: 0.4418 - val_accuracy: 0.7789
Epoch 6/20
40/40 - 0s - loss: 0.4079 - accuracy: 0.8017 - val_loss: 0.4385 - val_accuracy: 0.7821
Epoch 7/20
40/40 - 0s - loss: 0.4036 - accuracy: 0.8084 - val_loss: 0.4351 - val_accuracy: 0.7836
Epoch 8/20
40/40 - 0s - loss: 0.3943 - accuracy: 0.8110 - val_loss: 0.4372 - val_accuracy: 0.7836
Epoch 9/20
40/40 - 0s - loss: 0.3917 - accuracy: 0.8104 - val_loss: 0.4356 - val_accuracy: 0.7844
Epoch 10/20
40/40 - 0s - loss: 0.3897 - accuracy: 0.8092 - val_loss: 0.4311 - val_accuracy: 0.7876
Epoch 11/20
40/40 - 0s - loss: 0.3837 - accuracy: 0.8157 - val_loss: 0.4312 - val_accuracy: 0.7891
Epoch 12/20
40/40 - 0s - loss: 0.3817 - accuracy: 0.8159 - val_loss: 0.4314 - val_accuracy: 0.7907
Epoch 13/20
40/40 - 0s - loss: 0.3799 - accuracy: 0.8188 - val_loss: 0.4298 - val_accuracy: 0.7884
Epoch 14/20
40/40 - 0s - loss: 0.3750 - accuracy: 0.8220 - val_loss: 0.4326 - val_accuracy: 0.7884
Epoch 15/20
40/40 - 0s - loss: 0.3754 - accuracy: 0.8247 - val_loss: 0.4327 - val_accuracy: 0.7891
Epoch 16/20
40/40 - 0s - loss: 0.3703 - accuracy: 0.8265 - val_loss: 0.4316 - val_accuracy: 0.7946
Epoch 17/20
40/40 - 0s - loss: 0.3710 - accuracy: 0.8218 - val_loss: 0.4326 - val_accuracy: 0.7891
Epoch 18/20
40/40 - 0s - loss: 0.3677 - accuracy: 0.8257 - val_loss: 0.4318 - val_accuracy: 0.7891
Epoch 19/20
40/40 - 0s - loss: 0.3651 - accuracy: 0.8308 - val_loss: 0.4301 - val_accuracy: 0.7931
Epoch 20/20
40/40 - 0s - loss: 0.3615 - accuracy: 0.8293 - val_loss: 0.4293 - val_accuracy: 0.7946

plot1(history2)
../_images/dl-simple-case_68_0.png ../_images/dl-simple-case_68_1.png 
model2.evaluate(X_test, y_test, batch_size=128, verbose=2)

13/13 - 0s - loss: 0.4122 - accuracy: 0.8118

[0.4122048318386078, 0.8118313550949097]

Model 3: Regularization and Dropout

  • Based on the validation results of the previous two models, we can see that they are probably a bit overfit because the model performance on the validation set starts to stall after the first few epochs.

  • We can add regularization and dropouts in our network definition to avoid overfitting.

## Define embedding dimension
EMBEDDING_DIM = 128

## Define model
model3 = Sequential()
model3.add(
    Embedding(input_dim=vocab_size,
              output_dim=EMBEDDING_DIM,
              input_length=max_len,
              mask_zero=True))
model3.add(
    layers.SimpleRNN(16,
                     activation="relu",
                     name="RNN_layer",
                     dropout=0.2,
                     recurrent_dropout=0.2))  ## add dropout
model3.add(Dense(16, activation="relu", name="dense_layer"))
model3.add(Dense(1, activation="sigmoid", name="output"))

model3.compile(loss=keras.losses.BinaryCrossentropy(),
               optimizer=keras.optimizers.Adam(lr=0.001),
               metrics=["accuracy"])

plot_model(model3)
../_images/dl-simple-case_73_0.png 
history3 = model3.fit(X_train,
                      y_train,
                      batch_size=BATCH_SIZE,
                      epochs=EPOCHS,
                      verbose=2,
                      validation_split=VALIDATION_SPLIT)

Epoch 1/20
40/40 - 2s - loss: 0.6677 - accuracy: 0.6528 - val_loss: 0.6221 - val_accuracy: 0.7120
Epoch 2/20
40/40 - 0s - loss: 0.5499 - accuracy: 0.7295 - val_loss: 0.4968 - val_accuracy: 0.7679
Epoch 3/20
40/40 - 0s - loss: 0.4652 - accuracy: 0.7665 - val_loss: 0.4655 - val_accuracy: 0.7687
Epoch 4/20
40/40 - 0s - loss: 0.4496 - accuracy: 0.7683 - val_loss: 0.4566 - val_accuracy: 0.7703
Epoch 5/20
40/40 - 0s - loss: 0.4411 - accuracy: 0.7799 - val_loss: 0.4525 - val_accuracy: 0.7758
Epoch 6/20
40/40 - 0s - loss: 0.4365 - accuracy: 0.7860 - val_loss: 0.4501 - val_accuracy: 0.7710
Epoch 7/20
40/40 - 0s - loss: 0.4288 - accuracy: 0.7893 - val_loss: 0.4439 - val_accuracy: 0.7781
Epoch 8/20
40/40 - 0s - loss: 0.4273 - accuracy: 0.7954 - val_loss: 0.4446 - val_accuracy: 0.7718
Epoch 9/20
40/40 - 0s - loss: 0.4279 - accuracy: 0.7862 - val_loss: 0.4394 - val_accuracy: 0.7781
Epoch 10/20
40/40 - 0s - loss: 0.4214 - accuracy: 0.7941 - val_loss: 0.4377 - val_accuracy: 0.7742
Epoch 11/20
40/40 - 0s - loss: 0.4241 - accuracy: 0.7927 - val_loss: 0.4349 - val_accuracy: 0.7805
Epoch 12/20
40/40 - 0s - loss: 0.4153 - accuracy: 0.7950 - val_loss: 0.4358 - val_accuracy: 0.7821
Epoch 13/20
40/40 - 0s - loss: 0.4167 - accuracy: 0.7925 - val_loss: 0.4415 - val_accuracy: 0.7773
Epoch 14/20
40/40 - 0s - loss: 0.4138 - accuracy: 0.8033 - val_loss: 0.4335 - val_accuracy: 0.7797
Epoch 15/20
40/40 - 0s - loss: 0.4089 - accuracy: 0.7980 - val_loss: 0.4409 - val_accuracy: 0.7828
Epoch 16/20
40/40 - 0s - loss: 0.4138 - accuracy: 0.7958 - val_loss: 0.4355 - val_accuracy: 0.7797
Epoch 17/20
40/40 - 0s - loss: 0.4130 - accuracy: 0.7992 - val_loss: 0.4330 - val_accuracy: 0.7797
Epoch 18/20
40/40 - 0s - loss: 0.4106 - accuracy: 0.7966 - val_loss: 0.4332 - val_accuracy: 0.7836
Epoch 19/20
40/40 - 0s - loss: 0.4062 - accuracy: 0.7996 - val_loss: 0.4293 - val_accuracy: 0.7821
Epoch 20/20
40/40 - 0s - loss: 0.4140 - accuracy: 0.7945 - val_loss: 0.4298 - val_accuracy: 0.7734

plot1(history3)
../_images/dl-simple-case_75_0.png ../_images/dl-simple-case_75_1.png 
model3.evaluate(X_test, y_test, batch_size=128, verbose=2)

13/13 - 0s - loss: 0.4066 - accuracy: 0.8175

[0.40660879015922546, 0.8174952864646912]

Model 4: Improve the Models

  • In addition to regularization and dropouts, we can further improve the model by increasing the model complexity.

  • In particular, we can increase the depths and widths of the network layers.

  • Let’s try stacking two RNN layers.

Tip

When we stack two sequence layers (e.g., RNN), we need to make sure that the hidden states (outputs) of the first sequence layer at all timesteps are properly passed onto the next sequence layer, not just the hidden state (output) of the last timestep.

In keras, this usually means that we need to set the argument return_sequences=True in a sequence layer (e.g., SimpleRNN, LSTM, GRU etc).

## Define embedding dimension
MBEDDING_DIM = 128

## Define model
model4 = Sequential()
model4.add(
    Embedding(input_dim=vocab_size,
              output_dim=EMBEDDING_DIM,
              input_length=max_len,
              mask_zero=True))
model4.add(
    layers.SimpleRNN(16,
                     activation="relu",
                     name="RNN_layer_1",
                     dropout=0.2,
                     recurrent_dropout=0.5,
                     return_sequences=True)) ## To ensure the hidden states of all timesteps are pased down to next layer
model4.add(
    layers.SimpleRNN(16,
                     activation="relu",
                     name="RNN_layer_2",
                     dropout=0.2,
                     recurrent_dropout=0.5))
model4.add(Dense(1, activation="sigmoid", name="output"))

## Compile model
model4.compile(loss=keras.losses.BinaryCrossentropy(),
               optimizer=keras.optimizers.Adam(lr=0.001),
               metrics=["accuracy"])

plot_model(model4)
../_images/dl-simple-case_81_0.png 
history4 = model4.fit(X_train,
                      y_train,
                      batch_size=BATCH_SIZE,
                      epochs=EPOCHS,
                      verbose=2,
                      validation_split=VALIDATION_SPLIT)

Epoch 1/20
40/40 - 3s - loss: 0.6527 - accuracy: 0.6223 - val_loss: 0.6090 - val_accuracy: 0.6404
Epoch 2/20
40/40 - 0s - loss: 0.6050 - accuracy: 0.6277 - val_loss: 0.5654 - val_accuracy: 0.6491
Epoch 3/20
40/40 - 0s - loss: 0.5609 - accuracy: 0.6528 - val_loss: 0.5366 - val_accuracy: 0.7207
Epoch 4/20
40/40 - 0s - loss: 0.5317 - accuracy: 0.7061 - val_loss: 0.5238 - val_accuracy: 0.7168
Epoch 5/20
40/40 - 0s - loss: 0.5161 - accuracy: 0.7238 - val_loss: 0.5155 - val_accuracy: 0.7254
Epoch 6/20
40/40 - 0s - loss: 0.5157 - accuracy: 0.7352 - val_loss: 0.5135 - val_accuracy: 0.7223
Epoch 7/20
40/40 - 0s - loss: 0.5085 - accuracy: 0.7343 - val_loss: 0.5121 - val_accuracy: 0.7215
Epoch 8/20
40/40 - 0s - loss: 0.5004 - accuracy: 0.7500 - val_loss: 0.5101 - val_accuracy: 0.7238
Epoch 9/20
40/40 - 0s - loss: 0.4994 - accuracy: 0.7470 - val_loss: 0.5058 - val_accuracy: 0.7262
Epoch 10/20
40/40 - 0s - loss: 0.4930 - accuracy: 0.7563 - val_loss: 0.5032 - val_accuracy: 0.7325
Epoch 11/20
40/40 - 0s - loss: 0.4909 - accuracy: 0.7571 - val_loss: 0.4972 - val_accuracy: 0.7301
Epoch 12/20
40/40 - 0s - loss: 0.4846 - accuracy: 0.7573 - val_loss: 0.4938 - val_accuracy: 0.7317
Epoch 13/20
40/40 - 0s - loss: 0.4846 - accuracy: 0.7526 - val_loss: 0.4873 - val_accuracy: 0.7349
Epoch 14/20
40/40 - 0s - loss: 0.4815 - accuracy: 0.7508 - val_loss: 0.4841 - val_accuracy: 0.7317
Epoch 15/20
40/40 - 0s - loss: 0.4754 - accuracy: 0.7624 - val_loss: 0.4791 - val_accuracy: 0.7498
Epoch 16/20
40/40 - 0s - loss: 0.4648 - accuracy: 0.7689 - val_loss: 0.4749 - val_accuracy: 0.7522
Epoch 17/20
40/40 - 0s - loss: 0.4627 - accuracy: 0.7693 - val_loss: 0.4699 - val_accuracy: 0.7490
Epoch 18/20
40/40 - 0s - loss: 0.4682 - accuracy: 0.7649 - val_loss: 0.4673 - val_accuracy: 0.7553
Epoch 19/20
40/40 - 0s - loss: 0.4658 - accuracy: 0.7675 - val_loss: 0.4672 - val_accuracy: 0.7498
Epoch 20/20
40/40 - 0s - loss: 0.4656 - accuracy: 0.7655 - val_loss: 0.4666 - val_accuracy: 0.7514

plot1(history4)
../_images/dl-simple-case_83_0.png ../_images/dl-simple-case_83_1.png 
model4.evaluate(X_test, y_test, batch_size=128, verbose=2)

13/13 - 0s - loss: 0.4327 - accuracy: 0.7936

[0.43272581696510315, 0.7935808897018433]

Model 5: Bidirectional

  • We can also increase the model complexity in at least two possible ways:

    • Use more advanced RNNs, such as LSTM or GRU

    • Process the sequence in two directions

    • Increase the hidden nodes of the RNN/LSTM

  • Now let’s try the more sophisticated RNN, LSTM, and with bidirectional sequence processing and add more nodes to the LSTM layer.

## Define embedding dimension
EMBEDDING_DIM = 128

## Define model
model5 = Sequential()
model5.add(
    Embedding(input_dim=vocab_size,
              output_dim=EMBEDDING_DIM,
              input_length=max_len,
              mask_zero=True))
model5.add(
    layers.Bidirectional(  ## Bidirectional sequence processing
        LSTM(32,
             activation="relu",
             name="lstm_layer_1",
             dropout=0.2,
             recurrent_dropout=0.5,
             return_sequences=True)))
model5.add(
    layers.Bidirectional(  ## Bidirectional sequence processing
        LSTM(32,
             activation="relu",
             name="lstm_layer_2",
             dropout=0.2,
             recurrent_dropout=0.5)))
model5.add(Dense(1, activation="sigmoid", name="output"))

model5.compile(loss=keras.losses.BinaryCrossentropy(),
               optimizer=keras.optimizers.Adam(lr=0.001),
               metrics=["accuracy"])

plot_model(model5)
../_images/dl-simple-case_88_0.png 
history5 = model5.fit(X_train,
                      y_train,
                      batch_size=BATCH_SIZE,
                      epochs=EPOCHS,
                      verbose=2,
                      validation_split=VALIDATION_SPLIT)

Epoch 1/20
40/40 - 13s - loss: 0.6574 - accuracy: 0.6255 - val_loss: 0.6185 - val_accuracy: 0.6404
Epoch 2/20
40/40 - 2s - loss: 0.5829 - accuracy: 0.6629 - val_loss: 0.5311 - val_accuracy: 0.7301
Epoch 3/20
40/40 - 2s - loss: 0.5027 - accuracy: 0.7425 - val_loss: 0.4978 - val_accuracy: 0.7435
Epoch 4/20
40/40 - 2s - loss: 0.4677 - accuracy: 0.7693 - val_loss: 0.4870 - val_accuracy: 0.7655
Epoch 5/20
40/40 - 2s - loss: 0.4554 - accuracy: 0.7775 - val_loss: 0.4625 - val_accuracy: 0.7766
Epoch 6/20
40/40 - 2s - loss: 0.4391 - accuracy: 0.7901 - val_loss: 0.4546 - val_accuracy: 0.7805
Epoch 7/20
40/40 - 2s - loss: 0.4228 - accuracy: 0.7982 - val_loss: 0.4561 - val_accuracy: 0.7703
Epoch 8/20
40/40 - 2s - loss: 0.4253 - accuracy: 0.7988 - val_loss: 0.4408 - val_accuracy: 0.7844
Epoch 9/20
40/40 - 2s - loss: 0.4158 - accuracy: 0.8019 - val_loss: 0.4379 - val_accuracy: 0.7876
Epoch 10/20
40/40 - 2s - loss: 0.4162 - accuracy: 0.8049 - val_loss: 0.4329 - val_accuracy: 0.7899
Epoch 11/20
40/40 - 2s - loss: 0.4085 - accuracy: 0.8080 - val_loss: 0.4329 - val_accuracy: 0.7915
Epoch 12/20
40/40 - 2s - loss: 0.4051 - accuracy: 0.8070 - val_loss: 0.4294 - val_accuracy: 0.7923
Epoch 13/20
40/40 - 2s - loss: 0.4065 - accuracy: 0.8070 - val_loss: 0.4290 - val_accuracy: 0.7891
Epoch 14/20
40/40 - 2s - loss: 0.4014 - accuracy: 0.8084 - val_loss: 0.4282 - val_accuracy: 0.7876
Epoch 15/20
40/40 - 2s - loss: 0.3971 - accuracy: 0.8070 - val_loss: 0.4249 - val_accuracy: 0.7986
Epoch 16/20
40/40 - 2s - loss: 0.3986 - accuracy: 0.8096 - val_loss: 0.4215 - val_accuracy: 0.7962
Epoch 17/20
40/40 - 2s - loss: 0.3950 - accuracy: 0.8120 - val_loss: 0.4184 - val_accuracy: 0.7986
Epoch 18/20
40/40 - 2s - loss: 0.3891 - accuracy: 0.8106 - val_loss: 0.4202 - val_accuracy: 0.7978
Epoch 19/20
40/40 - 2s - loss: 0.3918 - accuracy: 0.8090 - val_loss: 0.4246 - val_accuracy: 0.7970
Epoch 20/20
40/40 - 2s - loss: 0.3850 - accuracy: 0.8116 - val_loss: 0.4234 - val_accuracy: 0.7946

plot1(history5)
../_images/dl-simple-case_90_0.png ../_images/dl-simple-case_90_1.png 
model5.evaluate(X_test, y_test, batch_size=128, verbose=2)

13/13 - 0s - loss: 0.4050 - accuracy: 0.8125

[0.40498316287994385, 0.8124606609344482]

Check Embeddings

  • Compared to one-hot encodings of characters, embeddings may include more information relating to the characteristics (semantics?) of the characters.

  • We can extract the embedding layer and apply dimensional reduction techniques (i.e., TSNE) to see how embeddings capture the relationships in-between characters.

## A name in sequence from test set
X_test[10]

array([ 0,  0,  0,  0,  0,  0,  0,  0,  9, 13,  2, 22,  1,  5, 10],
      dtype=int32)

ind2char = tokenizer.index_word
[ind2char.get(i) for i in X_test[10] if ind2char.get(i) != None]

['s', 'h', 'e', 'p', 'a', 'r', 'd']

tokenizer.texts_to_sequences('Alvin')

[[1], [6], [20], [3], [4]]

char_vectors = model5.layers[0].get_weights()[0]
char_vectors.shape

(30, 128)

labels = [char for (ind, char) in tokenizer.index_word.items()]
labels.insert(0, None)
labels

[None,
 'a',
 'e',
 'i',
 'n',
 'r',
 'l',
 'o',
 't',
 's',
 'd',
 'y',
 'm',
 'h',
 'c',
 'b',
 'u',
 'g',
 'k',
 'j',
 'v',
 'f',
 'p',
 'w',
 'z',
 'q',
 'x',
 '-',
 ' ',
 "'"]

from sklearn.manifold import TSNE

tsne = TSNE(n_components=2, random_state=0, n_iter=5000, perplexity=3)
np.set_printoptions(suppress=True)
T = tsne.fit_transform(char_vectors)
labels = labels

plt.figure(figsize=(10, 7), dpi=150)
plt.scatter(T[:, 0], T[:, 1], c='orange', edgecolors='r')
for label, x, y in zip(labels, T[:, 0], T[:, 1]):
    plt.annotate(label,
                 xy=(x + 1, y + 1),
                 xytext=(0, 0),
                 textcoords='offset points')
../_images/dl-simple-case_99_0.png

Issues of Word/Character Representations

  • One-hot encoding does not indicate semantic relationships between characters.

  • For deep learning NLP, it is preferred to convert one-hot encodings of words/characters into embeddings, which are argued to include more semantic information of the tokens.

  • Now the question is how to train and create better word embeddings. We will come back to this issue later.

Hyperparameter Tuning

Note

Please install keras tuner module in your current conda:

pip install -U keras-tuner

or

conda install -c conda-forge keras-tuner

  • Like feature-based ML methods, neural networks also come with many hyperparameters, which require default values.

  • Typical hyperparameters include:

    • Number of nodes for the layer

    • Learning Rates

  • We can utilize the module, keras-tuner, to fine-tune the hyperparameters (i.e., to find the values that optimize the model performance).

  • Steps for Keras Tuner

    • First, wrap the model definition in a function, which takes a single hp argument.

    • Inside this function, replace any value we want to tune with a call to hyperparameter sampling methods, e.g. hp.Int() or hp.Choice(). The function should return a compiled model.

    • Next, instantiate a tuner object specifying our optimization objective and other search parameters.

    • Finally, start the search with the search() method, which takes the same arguments as Model.fit() in keras.

    • When the search is over, we can retrieve the best model and a summary of the results from the tunner.

## confirm if the right kernel is being used
import sys
sys.executable

'/Users/Alvin/opt/anaconda3/envs/python-notes/bin/python'

import kerastuner

## Wrap model definition in a function
## and specify the parameters needed for tuning
# def build_model(hp):
#     model1 = keras.Sequential()
#     model1.add(keras.Input(shape=(max_len,)))
#     model1.add(layers.Dense(hp.Int('units', min_value=32, max_value=128, step=32), activation="relu", name="dense_layer_1"))
#     model1.add(layers.Dense(hp.Int('units', min_value=32, max_value=128, step=32), activation="relu", name="dense_layer_2"))
#     model1.add(layers.Dense(2, activation="softmax", name="output"))
#     model1.compile(
#         optimizer=keras.optimizers.Adam(
#             hp.Choice('learning_rate',
#                       values=[1e-2, 1e-3, 1e-4])),
#         loss='sparse_categorical_crossentropy',
#         metrics=['accuracy'])
#     return model1


def build_model(hp):
    m = Sequential()
    m.add(
        Embedding(input_dim=vocab_size,
                  output_dim=hp.Int('output_dim',  ## tuning 2
                                    min_value=32,
                                    max_value=128,
                                    step=32),
                  input_length=max_len,
                  mask_zero=True))
    m.add(
        layers.Bidirectional(
            LSTM(hp.Int('units', min_value=16, max_value=64, step=16), ## tuning 1
                 activation="relu",
                 dropout=0.2,
                 recurrent_dropout=0.2)))
    m.add(Dense(1, activation="sigmoid", name="output"))

    m.compile(loss=keras.losses.BinaryCrossentropy(),
              optimizer=keras.optimizers.Adam(lr=0.001),
              metrics=["accuracy"])
    return m

## This is to clean up the temp dir from the tuner
## Every time we re-start the tunner, it's better to keep the temp dir clean

import os
import shutil

if os.path.isdir('my_dir'):
    shutil.rmtree('my_dir')

  • The max_trials variable represents the maximum number of trials that a hyperparameter combination would run.

  • The execution_per_trial variable is the number of models that should be built and fit for each trial for robustness purposes.

## Instantiate the tunner

tuner = kerastuner.tuners.RandomSearch(build_model,
                                       objective='val_accuracy',
                                       max_trials=10,
                                       executions_per_trial=2,
                                       directory='my_dir')

## Check the tuner's search space
tuner.search_space_summary()

Search space summary

|-Default search space size: 2

output_dim (Int)

|-default: None
|-max_value: 128
|-min_value: 32
|-sampling: None
|-step: 32

units (Int)

|-default: None
|-max_value: 64
|-min_value: 16
|-sampling: None
|-step: 16
%%time
## Start tuning with the tuner
tuner.search(X_train, y_train, validation_split=0.2, batch_size=128)

40/40 [==============================] - ETA: 3:23 - loss: 0.6935 - accuracy: 0.48 - ETA: 1s - loss: 0.6927 - accuracy: 0.5408 - ETA: 0s - loss: 0.6916 - accuracy: 0.57 - ETA: 0s - loss: 0.6909 - accuracy: 0.57 - ETA: 0s - loss: 0.6901 - accuracy: 0.58 - ETA: 0s - loss: 0.6894 - accuracy: 0.58 - ETA: 0s - loss: 0.6887 - accuracy: 0.58 - ETA: 0s - loss: 0.6879 - accuracy: 0.59 - ETA: 0s - loss: 0.6870 - accuracy: 0.59 - ETA: 0s - loss: 0.6856 - accuracy: 0.59 - ETA: 0s - loss: 0.6847 - accuracy: 0.60 - ETA: 0s - loss: 0.6839 - accuracy: 0.60 - ETA: 0s - loss: 0.6831 - accuracy: 0.60 - ETA: 0s - loss: 0.6822 - accuracy: 0.60 - ETA: 0s - loss: 0.6810 - accuracy: 0.60 - ETA: 0s - loss: 0.6802 - accuracy: 0.60 - ETA: 0s - loss: 0.6789 - accuracy: 0.60 - ETA: 0s - loss: 0.6781 - accuracy: 0.61 - 7s 47ms/step - loss: 0.6777 - accuracy: 0.6103 - val_loss: 0.6219 - val_accuracy: 0.6404
40/40 [==============================] - ETA: 3:17 - loss: 0.6928 - accuracy: 0.48 - ETA: 1s - loss: 0.6912 - accuracy: 0.5651 - ETA: 0s - loss: 0.6900 - accuracy: 0.59 - ETA: 0s - loss: 0.6892 - accuracy: 0.60 - ETA: 0s - loss: 0.6878 - accuracy: 0.61 - ETA: 0s - loss: 0.6871 - accuracy: 0.61 - ETA: 0s - loss: 0.6863 - accuracy: 0.62 - ETA: 0s - loss: 0.6852 - accuracy: 0.62 - ETA: 0s - loss: 0.6845 - accuracy: 0.62 - ETA: 0s - loss: 0.6833 - accuracy: 0.62 - ETA: 0s - loss: 0.6827 - accuracy: 0.62 - ETA: 0s - loss: 0.6821 - accuracy: 0.62 - ETA: 0s - loss: 0.6815 - accuracy: 0.62 - ETA: 0s - loss: 0.6807 - accuracy: 0.62 - ETA: 0s - loss: 0.6799 - accuracy: 0.62 - ETA: 0s - loss: 0.6791 - accuracy: 0.62 - ETA: 0s - loss: 0.6784 - accuracy: 0.62 - ETA: 0s - loss: 0.6776 - accuracy: 0.62 - ETA: 0s - loss: 0.6768 - accuracy: 0.62 - 7s 45ms/step - loss: 0.6764 - accuracy: 0.6244 - val_loss: 0.6196 - val_accuracy: 0.6404

Trial complete

Trial summary

|-Trial ID: 91423e2fb669f90f41d5277b04f4f763
|-Score: 0.6404405832290649
|-Best step: 0

Hyperparameters:

|-output_dim: 64
|-units: 32
40/40 [==============================] - ETA: 3:16 - loss: 0.6952 - accuracy: 0.35 - ETA: 1s - loss: 0.6942 - accuracy: 0.4280 - ETA: 0s - loss: 0.6935 - accuracy: 0.46 - ETA: 0s - loss: 0.6927 - accuracy: 0.48 - ETA: 0s - loss: 0.6919 - accuracy: 0.50 - ETA: 0s - loss: 0.6911 - accuracy: 0.51 - ETA: 0s - loss: 0.6903 - accuracy: 0.52 - ETA: 0s - loss: 0.6895 - accuracy: 0.53 - ETA: 0s - loss: 0.6888 - accuracy: 0.54 - ETA: 0s - loss: 0.6879 - accuracy: 0.54 - ETA: 0s - loss: 0.6871 - accuracy: 0.55 - ETA: 0s - loss: 0.6862 - accuracy: 0.55 - ETA: 0s - loss: 0.6853 - accuracy: 0.56 - ETA: 0s - loss: 0.6844 - accuracy: 0.56 - ETA: 0s - loss: 0.6835 - accuracy: 0.56 - ETA: 0s - loss: 0.6824 - accuracy: 0.57 - ETA: 0s - loss: 0.6813 - accuracy: 0.57 - ETA: 0s - loss: 0.6803 - accuracy: 0.57 - ETA: 0s - loss: 0.6792 - accuracy: 0.57 - ETA: 0s - loss: 0.6782 - accuracy: 0.57 - 7s 47ms/step - loss: 0.6772 - accuracy: 0.5815 - val_loss: 0.6169 - val_accuracy: 0.6397
40/40 [==============================] - ETA: 3:17 - loss: 0.6921 - accuracy: 0.62 - ETA: 1s - loss: 0.6915 - accuracy: 0.6324 - ETA: 1s - loss: 0.6912 - accuracy: 0.62 - ETA: 1s - loss: 0.6908 - accuracy: 0.62 - ETA: 0s - loss: 0.6897 - accuracy: 0.61 - ETA: 1s - loss: 0.6894 - accuracy: 0.61 - ETA: 0s - loss: 0.6884 - accuracy: 0.61 - ETA: 0s - loss: 0.6878 - accuracy: 0.61 - ETA: 0s - loss: 0.6867 - accuracy: 0.61 - ETA: 0s - loss: 0.6856 - accuracy: 0.62 - ETA: 0s - loss: 0.6847 - accuracy: 0.62 - ETA: 0s - loss: 0.6837 - accuracy: 0.62 - ETA: 0s - loss: 0.6829 - accuracy: 0.62 - ETA: 0s - loss: 0.6820 - accuracy: 0.62 - ETA: 0s - loss: 0.6811 - accuracy: 0.62 - ETA: 0s - loss: 0.6801 - accuracy: 0.62 - ETA: 0s - loss: 0.6791 - accuracy: 0.62 - ETA: 0s - loss: 0.6782 - accuracy: 0.62 - ETA: 0s - loss: 0.6773 - accuracy: 0.62 - ETA: 0s - loss: 0.6764 - accuracy: 0.62 - ETA: 0s - loss: 0.6754 - accuracy: 0.62 - 7s 52ms/step - loss: 0.6744 - accuracy: 0.6255 - val_loss: 0.6133 - val_accuracy: 0.6397

Trial complete

Trial summary

|-Trial ID: 4b22b06e24c4dbcfbd2c21126b22dd93
|-Score: 0.6396538019180298
|-Best step: 0

Hyperparameters:

|-output_dim: 96
|-units: 32
40/40 [==============================] - ETA: 3:19 - loss: 0.6969 - accuracy: 0.38 - ETA: 1s - loss: 0.6960 - accuracy: 0.4045 - ETA: 1s - loss: 0.6948 - accuracy: 0.44 - ETA: 1s - loss: 0.6938 - accuracy: 0.46 - ETA: 0s - loss: 0.6928 - accuracy: 0.48 - ETA: 0s - loss: 0.6919 - accuracy: 0.50 - ETA: 0s - loss: 0.6910 - accuracy: 0.51 - ETA: 0s - loss: 0.6906 - accuracy: 0.51 - ETA: 0s - loss: 0.6897 - accuracy: 0.52 - ETA: 0s - loss: 0.6887 - accuracy: 0.52 - ETA: 0s - loss: 0.6878 - accuracy: 0.53 - ETA: 0s - loss: 0.6869 - accuracy: 0.53 - ETA: 0s - loss: 0.6860 - accuracy: 0.54 - ETA: 0s - loss: 0.6852 - accuracy: 0.54 - ETA: 0s - loss: 0.6841 - accuracy: 0.55 - ETA: 0s - loss: 0.6830 - accuracy: 0.55 - ETA: 0s - loss: 0.6824 - accuracy: 0.55 - ETA: 0s - loss: 0.6813 - accuracy: 0.55 - ETA: 0s - loss: 0.6802 - accuracy: 0.56 - ETA: 0s - loss: 0.6790 - accuracy: 0.56 - ETA: 0s - loss: 0.6778 - accuracy: 0.56 - 7s 53ms/step - loss: 0.6767 - accuracy: 0.5686 - val_loss: 0.6139 - val_accuracy: 0.6404
40/40 [==============================] - ETA: 3:17 - loss: 0.6937 - accuracy: 0.38 - ETA: 1s - loss: 0.6928 - accuracy: 0.4857 - ETA: 1s - loss: 0.6917 - accuracy: 0.52 - ETA: 1s - loss: 0.6906 - accuracy: 0.54 - ETA: 1s - loss: 0.6895 - accuracy: 0.56 - ETA: 1s - loss: 0.6884 - accuracy: 0.57 - ETA: 0s - loss: 0.6874 - accuracy: 0.57 - ETA: 0s - loss: 0.6862 - accuracy: 0.58 - ETA: 0s - loss: 0.6852 - accuracy: 0.58 - ETA: 0s - loss: 0.6848 - accuracy: 0.59 - ETA: 0s - loss: 0.6837 - accuracy: 0.59 - ETA: 0s - loss: 0.6832 - accuracy: 0.59 - ETA: 0s - loss: 0.6823 - accuracy: 0.59 - ETA: 0s - loss: 0.6813 - accuracy: 0.59 - ETA: 0s - loss: 0.6804 - accuracy: 0.60 - ETA: 0s - loss: 0.6795 - accuracy: 0.60 - ETA: 0s - loss: 0.6786 - accuracy: 0.60 - ETA: 0s - loss: 0.6778 - accuracy: 0.60 - ETA: 0s - loss: 0.6769 - accuracy: 0.60 - ETA: 0s - loss: 0.6759 - accuracy: 0.60 - ETA: 0s - loss: 0.6750 - accuracy: 0.60 - 7s 57ms/step - loss: 0.6740 - accuracy: 0.6072 - val_loss: 0.6092 - val_accuracy: 0.6397

Trial complete

Trial summary

|-Trial ID: d47a6c38d5966a5e77f8840501932a1d
|-Score: 0.6400471925735474
|-Best step: 0

Hyperparameters:

|-output_dim: 96
|-units: 48
40/40 [==============================] - ETA: 3:15 - loss: 0.6941 - accuracy: 0.39 - ETA: 1s - loss: 0.6932 - accuracy: 0.4744 - ETA: 1s - loss: 0.6921 - accuracy: 0.51 - ETA: 1s - loss: 0.6912 - accuracy: 0.53 - ETA: 0s - loss: 0.6904 - accuracy: 0.55 - ETA: 0s - loss: 0.6895 - accuracy: 0.55 - ETA: 0s - loss: 0.6887 - accuracy: 0.56 - ETA: 0s - loss: 0.6878 - accuracy: 0.56 - ETA: 0s - loss: 0.6868 - accuracy: 0.57 - ETA: 0s - loss: 0.6858 - accuracy: 0.57 - ETA: 0s - loss: 0.6848 - accuracy: 0.58 - ETA: 0s - loss: 0.6839 - accuracy: 0.58 - ETA: 0s - loss: 0.6834 - accuracy: 0.58 - ETA: 0s - loss: 0.6822 - accuracy: 0.58 - ETA: 0s - loss: 0.6810 - accuracy: 0.59 - ETA: 0s - loss: 0.6800 - accuracy: 0.59 - ETA: 0s - loss: 0.6790 - accuracy: 0.59 - ETA: 0s - loss: 0.6781 - accuracy: 0.59 - ETA: 0s - loss: 0.6772 - accuracy: 0.59 - ETA: 0s - loss: 0.6762 - accuracy: 0.59 - ETA: 0s - loss: 0.6753 - accuracy: 0.59 - 7s 52ms/step - loss: 0.6749 - accuracy: 0.5992 - val_loss: 0.6177 - val_accuracy: 0.6397
40/40 [==============================] - ETA: 3:16 - loss: 0.6944 - accuracy: 0.37 - ETA: 1s - loss: 0.6934 - accuracy: 0.4640 - ETA: 1s - loss: 0.6924 - accuracy: 0.50 - ETA: 0s - loss: 0.6916 - accuracy: 0.52 - ETA: 0s - loss: 0.6910 - accuracy: 0.53 - ETA: 0s - loss: 0.6903 - accuracy: 0.53 - ETA: 0s - loss: 0.6896 - accuracy: 0.54 - ETA: 0s - loss: 0.6888 - accuracy: 0.55 - ETA: 0s - loss: 0.6880 - accuracy: 0.55 - ETA: 0s - loss: 0.6871 - accuracy: 0.55 - ETA: 0s - loss: 0.6867 - accuracy: 0.56 - ETA: 0s - loss: 0.6862 - accuracy: 0.56 - ETA: 0s - loss: 0.6852 - accuracy: 0.56 - ETA: 0s - loss: 0.6842 - accuracy: 0.56 - ETA: 0s - loss: 0.6837 - accuracy: 0.57 - ETA: 0s - loss: 0.6832 - accuracy: 0.57 - ETA: 0s - loss: 0.6820 - accuracy: 0.57 - ETA: 0s - loss: 0.6810 - accuracy: 0.57 - ETA: 0s - loss: 0.6799 - accuracy: 0.57 - ETA: 0s - loss: 0.6789 - accuracy: 0.58 - ETA: 0s - loss: 0.6779 - accuracy: 0.58 - ETA: 0s - loss: 0.6768 - accuracy: 0.58 - 7s 52ms/step - loss: 0.6758 - accuracy: 0.5859 - val_loss: 0.6211 - val_accuracy: 0.6404

Trial complete

Trial summary

|-Trial ID: bbb31f86d25d233374c692062b5bafbb
|-Score: 0.6400471925735474
|-Best step: 0

Hyperparameters:

|-output_dim: 64
|-units: 48
40/40 [==============================] - ETA: 3:16 - loss: 0.6897 - accuracy: 0.59 - ETA: 0s - loss: 0.6885 - accuracy: 0.6176 - ETA: 0s - loss: 0.6871 - accuracy: 0.62 - ETA: 0s - loss: 0.6863 - accuracy: 0.62 - ETA: 0s - loss: 0.6849 - accuracy: 0.62 - ETA: 0s - loss: 0.6838 - accuracy: 0.62 - ETA: 0s - loss: 0.6821 - accuracy: 0.63 - ETA: 0s - loss: 0.6811 - accuracy: 0.63 - ETA: 0s - loss: 0.6798 - accuracy: 0.63 - ETA: 0s - loss: 0.6789 - accuracy: 0.63 - ETA: 0s - loss: 0.6781 - accuracy: 0.63 - ETA: 0s - loss: 0.6773 - accuracy: 0.63 - ETA: 0s - loss: 0.6765 - accuracy: 0.63 - ETA: 0s - loss: 0.6753 - accuracy: 0.63 - ETA: 0s - loss: 0.6745 - accuracy: 0.63 - ETA: 0s - loss: 0.6733 - accuracy: 0.63 - ETA: 0s - loss: 0.6724 - accuracy: 0.63 - 7s 42ms/step - loss: 0.6717 - accuracy: 0.6301 - val_loss: 0.6167 - val_accuracy: 0.6404
40/40 [==============================] - ETA: 3:17 - loss: 0.6930 - accuracy: 0.52 - ETA: 1s - loss: 0.6917 - accuracy: 0.5747 - ETA: 0s - loss: 0.6906 - accuracy: 0.59 - ETA: 0s - loss: 0.6899 - accuracy: 0.60 - ETA: 0s - loss: 0.6893 - accuracy: 0.60 - ETA: 0s - loss: 0.6886 - accuracy: 0.60 - ETA: 0s - loss: 0.6878 - accuracy: 0.60 - ETA: 0s - loss: 0.6870 - accuracy: 0.60 - ETA: 0s - loss: 0.6861 - accuracy: 0.60 - ETA: 0s - loss: 0.6852 - accuracy: 0.61 - ETA: 0s - loss: 0.6844 - accuracy: 0.61 - ETA: 0s - loss: 0.6831 - accuracy: 0.61 - ETA: 0s - loss: 0.6822 - accuracy: 0.61 - ETA: 0s - loss: 0.6809 - accuracy: 0.61 - ETA: 0s - loss: 0.6799 - accuracy: 0.61 - ETA: 0s - loss: 0.6789 - accuracy: 0.61 - ETA: 0s - loss: 0.6779 - accuracy: 0.61 - ETA: 0s - loss: 0.6769 - accuracy: 0.61 - ETA: 0s - loss: 0.6759 - accuracy: 0.61 - 7s 45ms/step - loss: 0.6750 - accuracy: 0.6169 - val_loss: 0.6152 - val_accuracy: 0.6404

Trial complete

Trial summary

|-Trial ID: 19bb395cd66cbfbae6b62d81305bfd28
|-Score: 0.6404405832290649
|-Best step: 0

Hyperparameters:

|-output_dim: 96
|-units: 16
40/40 [==============================] - ETA: 3:16 - loss: 0.6937 - accuracy: 0.39 - ETA: 1s - loss: 0.6935 - accuracy: 0.4627 - ETA: 0s - loss: 0.6930 - accuracy: 0.49 - ETA: 0s - loss: 0.6925 - accuracy: 0.51 - ETA: 0s - loss: 0.6921 - accuracy: 0.52 - ETA: 0s - loss: 0.6916 - accuracy: 0.53 - ETA: 0s - loss: 0.6910 - accuracy: 0.54 - ETA: 0s - loss: 0.6901 - accuracy: 0.55 - ETA: 0s - loss: 0.6895 - accuracy: 0.55 - ETA: 0s - loss: 0.6889 - accuracy: 0.56 - ETA: 0s - loss: 0.6879 - accuracy: 0.56 - ETA: 0s - loss: 0.6873 - accuracy: 0.57 - ETA: 0s - loss: 0.6866 - accuracy: 0.57 - ETA: 0s - loss: 0.6858 - accuracy: 0.57 - ETA: 0s - loss: 0.6850 - accuracy: 0.57 - ETA: 0s - loss: 0.6843 - accuracy: 0.58 - ETA: 0s - loss: 0.6835 - accuracy: 0.58 - ETA: 0s - loss: 0.6828 - accuracy: 0.58 - ETA: 0s - loss: 0.6820 - accuracy: 0.58 - 7s 45ms/step - loss: 0.6813 - accuracy: 0.5880 - val_loss: 0.6360 - val_accuracy: 0.6404
40/40 [==============================] - ETA: 3:23 - loss: 0.6942 - accuracy: 0.38 - ETA: 1s - loss: 0.6935 - accuracy: 0.4492 - ETA: 0s - loss: 0.6929 - accuracy: 0.48 - ETA: 0s - loss: 0.6924 - accuracy: 0.50 - ETA: 0s - loss: 0.6918 - accuracy: 0.52 - ETA: 0s - loss: 0.6911 - accuracy: 0.53 - ETA: 0s - loss: 0.6904 - accuracy: 0.54 - ETA: 0s - loss: 0.6897 - accuracy: 0.55 - ETA: 0s - loss: 0.6891 - accuracy: 0.56 - ETA: 0s - loss: 0.6884 - accuracy: 0.56 - ETA: 0s - loss: 0.6878 - accuracy: 0.57 - ETA: 0s - loss: 0.6872 - accuracy: 0.57 - ETA: 0s - loss: 0.6866 - accuracy: 0.57 - ETA: 0s - loss: 0.6859 - accuracy: 0.58 - ETA: 0s - loss: 0.6852 - accuracy: 0.58 - ETA: 0s - loss: 0.6846 - accuracy: 0.58 - ETA: 0s - loss: 0.6839 - accuracy: 0.58 - ETA: 0s - loss: 0.6832 - accuracy: 0.58 - ETA: 0s - loss: 0.6825 - accuracy: 0.58 - ETA: 0s - loss: 0.6818 - accuracy: 0.59 - 7s 49ms/step - loss: 0.6812 - accuracy: 0.5924 - val_loss: 0.6327 - val_accuracy: 0.6404

Trial complete

Trial summary

|-Trial ID: c3453702db19c408ba1556f4df699223
|-Score: 0.6404405832290649
|-Best step: 0

Hyperparameters:

|-output_dim: 32
|-units: 48
INFO:tensorflow:Oracle triggered exit
CPU times: user 1min 54s, sys: 2.85 s, total: 1min 57s
Wall time: 1min 32s

## Retrieve the best models from the tuner
models = tuner.get_best_models(num_models=2)

plot_model(models[0], show_shapes=True)
../_images/dl-simple-case_115_0.png 
## Retrieve the summary of results from the tuner
tuner.results_summary()

Results summary

|-Results in my_dir/untitled_project
|-Showing 10 best trials
|-Objective(name='val_accuracy', direction='max')

Trial summary

|-Trial ID: 91423e2fb669f90f41d5277b04f4f763
|-Score: 0.6404405832290649
|-Best step: 0

Hyperparameters:

|-output_dim: 64
|-units: 32

Trial summary

|-Trial ID: 19bb395cd66cbfbae6b62d81305bfd28
|-Score: 0.6404405832290649
|-Best step: 0

Hyperparameters:

|-output_dim: 96
|-units: 16

Trial summary

|-Trial ID: c3453702db19c408ba1556f4df699223
|-Score: 0.6404405832290649
|-Best step: 0

Hyperparameters:

|-output_dim: 32
|-units: 48

Trial summary

|-Trial ID: d47a6c38d5966a5e77f8840501932a1d
|-Score: 0.6400471925735474
|-Best step: 0

Hyperparameters:

|-output_dim: 96
|-units: 48

Trial summary

|-Trial ID: bbb31f86d25d233374c692062b5bafbb
|-Score: 0.6400471925735474
|-Best step: 0

Hyperparameters:

|-output_dim: 64
|-units: 48

Trial summary

|-Trial ID: 4b22b06e24c4dbcfbd2c21126b22dd93
|-Score: 0.6396538019180298
|-Best step: 0

Hyperparameters:

|-output_dim: 96
|-units: 32

Explanation

Train Model with the Tuned Hyperparameters

EMBEDDING_DIM = 64
model6 = Sequential()
model6.add(
    Embedding(input_dim=vocab_size,
              output_dim=EMBEDDING_DIM,
              input_length=max_len,
              mask_zero=True))
model6.add(
    layers.Bidirectional(
        LSTM(64,
             activation="relu",
             name="lstm_layer",
             dropout=0.2,
             recurrent_dropout=0.5)))
model6.add(Dense(1, activation="sigmoid", name="output"))

model6.compile(loss=keras.losses.BinaryCrossentropy(),
               optimizer=keras.optimizers.Adam(lr=0.001),
               metrics=["accuracy"])
plot_model(model6)
../_images/dl-simple-case_119_0.png 
history6 = model6.fit(X_train,
                      y_train,
                      batch_size=BATCH_SIZE,
                      epochs=EPOCHS,
                      verbose=2,
                      validation_split=VALIDATION_SPLIT)

Epoch 1/20
40/40 - 8s - loss: 0.6570 - accuracy: 0.6261 - val_loss: 0.6158 - val_accuracy: 0.6404
Epoch 2/20
40/40 - 1s - loss: 0.5823 - accuracy: 0.6725 - val_loss: 0.5214 - val_accuracy: 0.7309
Epoch 3/20
40/40 - 1s - loss: 0.4951 - accuracy: 0.7610 - val_loss: 0.4929 - val_accuracy: 0.7640
Epoch 4/20
40/40 - 1s - loss: 0.4545 - accuracy: 0.7777 - val_loss: 0.4649 - val_accuracy: 0.7616
Epoch 5/20
40/40 - 1s - loss: 0.4419 - accuracy: 0.7856 - val_loss: 0.4566 - val_accuracy: 0.7703
Epoch 6/20
40/40 - 2s - loss: 0.4346 - accuracy: 0.7886 - val_loss: 0.4477 - val_accuracy: 0.7805
Epoch 7/20
40/40 - 1s - loss: 0.4276 - accuracy: 0.7945 - val_loss: 0.4450 - val_accuracy: 0.7821
Epoch 8/20
40/40 - 1s - loss: 0.4223 - accuracy: 0.7948 - val_loss: 0.4409 - val_accuracy: 0.7860
Epoch 9/20
40/40 - 1s - loss: 0.4196 - accuracy: 0.7988 - val_loss: 0.4349 - val_accuracy: 0.7844
Epoch 10/20
40/40 - 2s - loss: 0.4132 - accuracy: 0.8015 - val_loss: 0.4356 - val_accuracy: 0.7821
Epoch 11/20
40/40 - 2s - loss: 0.4099 - accuracy: 0.8035 - val_loss: 0.4314 - val_accuracy: 0.7907
Epoch 12/20
40/40 - 2s - loss: 0.4126 - accuracy: 0.8043 - val_loss: 0.4247 - val_accuracy: 0.7923
Epoch 13/20
40/40 - 1s - loss: 0.4032 - accuracy: 0.8082 - val_loss: 0.4231 - val_accuracy: 0.7954
Epoch 14/20
40/40 - 1s - loss: 0.4011 - accuracy: 0.8078 - val_loss: 0.4208 - val_accuracy: 0.7939
Epoch 15/20
40/40 - 2s - loss: 0.4008 - accuracy: 0.8080 - val_loss: 0.4181 - val_accuracy: 0.7970
Epoch 16/20
40/40 - 1s - loss: 0.3980 - accuracy: 0.8080 - val_loss: 0.4150 - val_accuracy: 0.7994
Epoch 17/20
40/40 - 1s - loss: 0.3968 - accuracy: 0.8143 - val_loss: 0.4206 - val_accuracy: 0.8041
Epoch 18/20
40/40 - 1s - loss: 0.3935 - accuracy: 0.8051 - val_loss: 0.4145 - val_accuracy: 0.8002
Epoch 19/20
40/40 - 1s - loss: 0.3957 - accuracy: 0.8122 - val_loss: 0.4144 - val_accuracy: 0.8049
Epoch 20/20
40/40 - 2s - loss: 0.3942 - accuracy: 0.8106 - val_loss: 0.4107 - val_accuracy: 0.8072

plot2(history6)
../_images/dl-simple-case_121_0.png 
from lime.lime_text import LimeTextExplainer

explainer = LimeTextExplainer(class_names=['male'], char_level=True)

def model_predict_pipeline(text):
    _seq = tokenizer.texts_to_sequences(text)
    _seq_pad = keras.preprocessing.sequence.pad_sequences(_seq, maxlen=max_len)
    #return np.array([[float(1-x), float(x)] for x in model.predict(np.array(_seq_pad))])
    return model6.predict(np.array(_seq_pad))


# np.array(sequence.pad_sequences(
#     tokenizer.texts_to_sequences([n for (n,l) in test_set]),
#     maxlen = max_len)).astype('float32')

reversed_word_index = dict([(index, word)
                            for (word, index) in tokenizer.word_index.items()])

text_id = 305

X_test[text_id]

array([ 0,  0,  0,  0,  0,  0,  0,  0,  0,  0, 18,  3, 12, 12,  7],
      dtype=int32)

X_test_texts[text_id]

'Kimmo'

' '.join([reversed_word_index.get(i, '?') for i in X_test[text_id]])

'? ? ? ? ? ? ? ? ? ? k i m m o'

model_predict_pipeline([X_test_texts[text_id]])

array([[0.80037665]], dtype=float32)

exp = explainer.explain_instance(X_test_texts[text_id],
                                 model_predict_pipeline,
                                 num_features=100,
                                 top_labels=1)

exp.show_in_notebook(text=True)

y_test[text_id]

1

exp = explainer.explain_instance('Tim',
                                 model_predict_pipeline,
                                 num_features=100,
                                 top_labels=1)
exp.show_in_notebook(text=True)

exp = explainer.explain_instance('Michaelis',
                                 model_predict_pipeline,
                                 num_features=100,
                                 top_labels=1)
exp.show_in_notebook(text=True)

exp = explainer.explain_instance('Sidney',
                                 model_predict_pipeline,
                                 num_features=100,
                                 top_labels=1)
exp.show_in_notebook(text=True)

exp = explainer.explain_instance('Timber',
                                 model_predict_pipeline,
                                 num_features=100,
                                 top_labels=1)
exp.show_in_notebook(text=True)

exp = explainer.explain_instance('Alvin',
                                 model_predict_pipeline,
                                 num_features=100,
                                 top_labels=1)
exp.show_in_notebook(text=True)

References

  • Chollet (2017), Ch 3 and Ch 4

Neural Network From Scratch Deep Learning: Sentiment Analysis